乌鸦传媒

By way of two resoundingly consistent votes, members of an FDA advisory committee agreed that PD-1 inhibitors do not have a favorable risk/benefit profile for gastroesophageal cancers with low PD-L1 expression <1.

The Oncologic Drugs Advisory Committee (ODAC) decided 10-2 with one abstention that PD-1 inhibitors do not have a favorable risk/benefit profile in gastric/gastroesophageal junction (GEJ) cancers with PD-L1 expression <1. Following a separate discussion, the panel agreed by an 11-1-1 vote on the same scenario in esophageal cancer, although panelists acknowledged difficulty in making sense of small subgroup analyses.

The day-long meeting came about as a result of new data that have become available since FDA granted PD-L1-agnostic approval for pembrolizumab (Keytruda) and nivolumab (Opdivo) as first-line treatment for advanced gastroesophageal cancers. BeiGene has a pending application for the same indications for tislelizumab (Tevimbra).

"I think there's some consistency in all the evidence that we've been presented," said acting ODAC chair Christopher Lieu, MD, of the University of Colorado Cancer Center in Aurora, at the end of the gastric/GEJ discussion. "I think PD-L1 is predictive of response. I think we see significant activity at PD-L1 greater than 10. I think we see some modest benefit between 1 and 10, and I've seen no evidence of benefit in PD-L1 scores less than 1."

The comments mirrored those of the vast majority of panelists after hearing presentations by FDA staff and the three sponsors during both sessions of the meeting. FDA staff acknowledged that the sponsors had used different methods to calculate PD-L1 expression and different cutoffs for response, as well as the hazards of post-hoc subgroup analyses. Lieu also noted reports of significant inter-observer variability in the interpretation and reporting of PD-L1 measurements.

Prevailing Uncertainty

Panelist Ravi Madan, MD, of the National Cancer Institute in Bethesda, Maryland, posted the lone abstention on both votes and captured the uncertainty expressed by multiple members of ODAC.

"I'm just not fully convinced that this is the dataset that should be used to address the question of, 'Is this the cutoff required to bring about benefit versus risk?'" he said. "We don't know. We never really went into these studies asking this question. That's where I have some trouble; that is, I'm not sure that we have the power in this dataset. We're trying our best to glean what we can from existing data that was never really designed to answer this question."

The gastric/GEJ discussion focused on data from three randomized trials of chemotherapy with or without PD-1 inhibition as first-line therapy for unresectable/metastatic disease. All three trials met prespecified endpoints for improved overall survival (OS).

The FDA presented a breakdown of the OS results by all patients and by cutoffs of PD-L1 <1 and PD-L1 <10. The all-comer analysis yielded statistically significant hazard ratios of 0.77-0.80. The PD-L1 <10 analysis showed smaller hazard reductions, and only the KEYNOTE-859 trial with pembrolizumab achieved statistical significance. The analysis of results with PD-L1 <1 ranged from 0.92-0.98, none of which achieved statistical significance.

The same breakdown for three randomized trials in esophageal cancer showed similar results: significant improvement in the overall populations, variable improvement with different PD-L1 cutoffs, and no evidence of improvement with a cutoff of PD-L1 <1. In fact, results from the trial with tislelizumab yielded a hazard ratio of 1.34 for chemotherapy plus PD-1 inhibition in cancers with PD-L1 <1, suggesting a detrimental effect with the treatment.

Several ODAC panelists expressed concern about the potential for introducing additional toxicity to patients who are already seriously ill, with little or no evidence of benefit.

FDA staff concluded its morning presentation by acknowledging that PD-L1 expression "appears to be a predictive biomarker" in gastric/GEJ cancer, but the data for PD-L1 <1 provided reason for concern about limited benefit with potentially more toxicity. The agency presentation during the esophageal cancer session ended with a summary stating the data consistently showed that "a distinct treatment effect emerges based on PD-L1 status." Use of immune checkpoint inhibitors in the setting of PD-L1 <1 may expose patients to toxicity with no clear evidence of benefit.

If FDA were to establish a cutoff of PD-L1 <1 -- which was not under consideration by ODAC -- 10-20% of patients with advanced gastric/GEJ and 10% or fewer patients with advanced esophageal cancer would not qualify to receive PD-1 inhibitors.

Merck officials most stridently supported the findings of their studies and concluded that the PD-L1-agnostic approval status should remain. Representatives for Bristol Myers Squibb outlined several potential approaches to the cutoff issue, but stated that maintaining the PD-L1-agnostic status "would be preferable." With their supplemental approval still pending, BeiGene officials seemed most open to discussing PD-L1 cutoffs for tislelizumab labeling.

Harm Versus Risk

During the esophageal cancer discussion, Merck vice president Catherine Pietanza, MD, pointed out that four patients with PD-L1 expression <1 had complete responses with pembrolizumab versus none with chemotherapy alone. Three of the four derived long-term benefit from the treatment, the implication being that patients with little or no PD-L1 expression may still benefit from the treatment.

ODAC patient representative Dana Deighton, who cast the lone positive vote during the session, expressed concern about withholding potentially useful treatment for patients who have little hope for extended survival.

"I believe in the Hail Mary pass," she said. "Sometimes things happen, and you don't know why."

Panelist Daniel Spratt, MD, of University Hospitals Seidman Cancer Center-Case Western Reserve in Cleveland, reminded the committee of the purpose of the two sessions.

"The question is not whether we should impose some restriction on cutpoint," he said. "The question is, does the cumulative data support the use of PD-L1 expression as a predictive biomarker at its core. There is a different relative effect size by biomarker status. Period. End of story."

"A point that I think gets confused a lot ... is just because a patient has an amazing response to chemo and IO [immuno-oncology] doesn't mean they wouldn't have an amazing response to chemo," he added. "With a hazard ratio of almost 1, they probably had a similar probability of having that benefit."

While the FDA is not required to follow the advice of its advisory committees, it typically does.

Comment